Eldric is the multi-tenant AI infrastructure platform behind ISP, MSP, telco, and regional-cloud Private AI offerings. Your customers want AI; they do not want to send their data to OpenAI, Anthropic, Azure, or AWS. With Eldric, you run the platform on your own hardware in your own datacenters — every customer is a fully-isolated tenant, fully branded as your service, billed by you.
You keep the customer relationship. You keep the recurring revenue. You keep the data inside the jurisdiction it belongs to. The thing you do not have to build is the AI platform itself.
Your enterprise customers are being asked, in board meetings, what their AI strategy is. They are answering the question one of three ways. Two of those answers leave you on the sideline — either watching the customer ship workloads to a hyperscaler, or watching them disappear into an eighteen-month internal build that may never finish. The third answer puts you at the centre of the relationship and gives you a new recurring revenue line that did not exist last year.
A Private-AI offering only works if the provider can confidently sell to enterprise customers who care about isolation, branding, billing, and where the bytes actually live. Eldric ships those four properties as platform primitives, not professional-services line items. The provider builds a sales motion; Eldric is what they sell.
Every persisted artefact carries a tenant ID. The platform refuses any cross-tenant access by default; a superadmin escape hatch exists for legitimate cluster-internal calls. Per-tenant role-based access, quotas, and capability tokens — the same RBAC the provider uses for their own admins.
Each tenant gets its own portal at its own subdomain, with its own colour palette, logo, typography, and copy. The chat shell, admin surfaces, and customer-facing UI all theme per tenant. The customer's users never need to know whose infrastructure is underneath.
The platform emits per-tenant usage events — requests, tokens, retrieval calls, training minutes, storage footprint — via OTLP-HTTP into whatever observability or billing stack the provider already runs. Turnkey connectors for common billing platforms ship as plugins on the roadmap; the raw signal is available today.
Pin a tenant to a region. The platform refuses to replicate that tenant's data outside the regions it is allowed to live in — a rule the cluster enforces, not the operator. Federation across the provider's sites is code-complete; the customer-visible dashboard for region pinning is the final a later 5.0.x patch polish.
If your customer is comparing your branded Private-AI offering with what they could get from a public cloud, the answer needs to be: the same capabilities, in their language, on your jurisdiction, billed in their currency. Eldric covers the long tail of what enterprises actually ask for — not just chat, but the agentic and domain-specific tooling that separates a serious offering from a demo.
Each capability is gated by license tier — the provider chooses which capabilities to expose to which tenant, with what quota, at what price. A small-business tenant might see chat and RAG only; an enterprise tenant might get the full domain-specific catalogue plus fine-tuning. The provider sets the menu.
Most service providers operate across more than one site — different countries, different regulatory regions, different colocation contracts. Eldric federates across those sites without forcing one giant cluster spanning the WAN. Each region keeps its own autonomous Eldric cluster for hot paths; a federation layer keeps directory, tenant policy, and resource provisioning consistent across the whole platform. Customers see one product. Providers see one dashboard. Each region keeps the bytes that are supposed to stay there.
The provider sees one platform. The customer sees their own brand. The regulator sees data that never crossed a border it was not allowed to cross.
Eldric ships with five license tiers. As a service provider, you license Eldric at the Enterprise or Custom tier and resell the capability bundle to your customers as your own SKUs, at your own retail price. Eldric does not contract with your customers; the provider keeps the relationship end to end. The map below is the menu of capability bundles you can build your SKUs from.
| Tier | Scope | What the bundle includes |
|---|---|---|
| Free | 1 controller · 2 workers | The trial bundle. A way to let a prospective customer experience the platform before they commit. Provider-branded; same isolation rules apply. |
| Standard | 3 workers · RAG + agents | Small-business customer. Chat, RAG over private documents, basic agents, custom models. The everyday Private-AI offering. |
| Professional | 10 workers · multi-site · metrics | Mid-market enterprise. Adds dashboard, metrics, security hardening, distributed RAG, training data generation. The "real" deployment SKU. |
| Enterprise | 50 workers · HA · orchestration | Large enterprise customer. Adds high-availability features, federation, full orchestration, AI-driven decisions, priority support. Full capability surface. |
| Custom | unlimited · everything | The provider's own license. Unlimited controllers, routers, workers across all your regions. This is what you operate the platform on. |
Licenses are signed offline with Ed25519. The platform validates them locally; no phone-home, no per-token usage feed leaving your premises. The provider operates everything inside their own perimeter — the licensing is something you pay us for once a year, not a runtime dependency on something we control.
Service providers do not get to throw out their existing operations stack to add a new product. Eldric is built to integrate — through a plugin host that runs server-side and client-side extensions — with what you already have: your network management system, your fleet of BMCs, your identity provider, your billing platform, your ticketing queue. The plugin host is shipping; the ecosystem of connectors is growing each release.
SNMP and syslog plugins for monitoring integration with the NMS the operations team already lives in — feeding Eldric platform health into the same dashboard they watch for everything else.
The plugin host can run Redfish-style integrations against the bare-metal fleet — useful for AI-driven incident triage, predictive maintenance, and cluster-wide power/thermal awareness.
Per-tenant identity provider configuration. The customer's users authenticate against the customer's existing directory; the provider does not become an identity broker by accident.
Per-tenant telemetry via OTLP-HTTP feeds the billing platform you already run. Turnkey connectors to common platforms (Zuora, Chargebee, custom invoice runs) ship as plugins on the roadmap; the raw signal is available today.
Tenant provisioning and de-provisioning hooks. When a customer signs a contract in the provider's CRM, an Eldric tenant gets stood up; when they churn, it gets archived. The provider's portal does not need to learn a new vocabulary.
The provider's support team gets per-tenant logs, audit trails, and usage history — surfaced through whatever ticketing system they already operate. The escalation path stays familiar.
Six steps separate the day a provider commits to launching a Private-AI offering from the day their first customer pays an invoice for it. The platform side — steps one through three — is the part Eldric is responsible for, and the part where we work alongside the provider. The customer-facing side — steps four through six — is what the platform was built to make routine, not a bespoke integration each time.
Single rack or multi-site, on the hardware the provider already operates. dnf install eldric-aios on the cluster nodes; license file activated; controller online.
Custom-tier license issued by Eldric covers the provider's whole estate. Capability tiers configured as the provider's customer SKUs.
Default theme, billing connector, customer-portal handoff, identity-provider hooks, support workflow. Done once; carries every future tenant.
Customer hits the provider's portal, picks a SKU, signs the contract. The portal calls Eldric; a new tenant is provisioned with the customer's branding and limits.
The customer's users land on the provider-branded portal at the customer's own subdomain. Chat, RAG, agents, fine-tunes — whatever the SKU includes. The word "Eldric" never appears unless the provider wants it to.
Usage events flow through the billing connector into the provider's existing invoice run. The customer pays the provider on the provider's normal billing schedule. The recurring revenue line is open.
Not every service provider should sell Private AI; not every customer base wants it. We would rather have the honest conversation up front than land a deal that does not work for either of us. Here is the cut.
Most service-provider conversations start with a thirty-minute call: your customer base, your existing infrastructure, your timeline, your competitive landscape. We will tell you whether Eldric is the right tool for what you want to do — and if it is, what the path from today to your first paying tenant looks like.
We are a small Austrian company. The person you talk to is the person who knows the codebase. The conversation is in English or German; the licensing is signed in Vienna; the platform runs wherever you put the hardware.